In this work, we fabricated high-voltage (HV) amorphous InGaZnO (a-IGZO) thin film transistors (TFTs) consisting of an ultra-thin buried layer of amorphous InZnO (a-IZO) and a drain offset region. With a drain offset length ( L _offset ) of 2 μm, a breakdown voltage ( V _BD ) of 457 V is achieved. The output current is four times higher than the conventional device, thanks to the a-IZO buried layer. TCAD simulation is performed to reveal the operating mechanism of the fabricated device. A sub-channel is formed at the a-IZO/a-IGZO interface to improve the current capability of the HV device, achieving an excellent specific on-resistance ( R _{on, sp} ) of 6.25×10 ³ mΩ∙cm ² and a Baliga’s Figure of Merit (BFOM) of 33.4 KW/cm ² .

The output characteristics of the insulated gate bipolar transistor (IGBT) are the critical metric for the measurement of power control and conversion of power electronic systems. Existing methods are characterized by potential issues, such as high cost and extremely low simulation efficiency. In this article, by combining compact models and artificial neural networks (ANNs), we propose a novel technique for predicting the static output characteristics of IGBTs. The proposed method can rapidly predict the Hefner static model parameters including performance parameters. By introducing the model parameters into the electronic design automation (EDA) circuit simulator, output characteristics can be obtained. In addition, a phased prediction (PP) scheme is proposed to further reduce the prediction error of the model parameters. The effectiveness of the method is verified by comparing the results of the proposed method with those in the technical computer-aided design (TCAD) simulation and datasheet. Meanwhile, the method significantly improves the speed compared with TCAD simulation, which can improve the efficiency of obtaining electrical characteristics and reduce the design cost.

The breakdown voltage model of Si-substrated AlGaN/GaN HEMTs with gate and drain field plates is proposed in this work. The silicon substrate and GaN buffer are considered as the depletion region in the modeling process. The analytical model shows great simplicity and veracity. It gives physical insights into the breakdown characteristics of the AlGaN/GaN HEMTs. The avalanche breakdown occurs at the field plate’s edge in lateral structure or at the interface in vertical structure. According to the analytical model, the relationship between the lateral breakdown and the vertical breakdown is demonstrated in this work. The breakdown location is affected by the variation of the structure parameters, including gate-to-drain distance, buffer thickness and field plates’ lengths. With the aid of the analytical model, effective guidance for device optimization to achieve high performance can be obtained.

The AlGaN/GaN high electron mobility transistor with a step–doped channel (SDC–HEMT) is first proposed in this paper. The potential distribution and the electric field (E–field) distribution of the device are explored by the numerical approach and analytical approach simultaneously. By introducing extra dopants to the channel layer, the E–field distribution along the AlGaN/GaN heterojunction interface is reshaped, resulting in an improved breakdown characteristic. An optimized doping concentration gradient of channel layer of 2 × 10¹⁶ cm⁻³/step is proposed and verified by simulations. The breakdown voltage (BV) of the optimized SDC–HEMT reaches 1486 V with a 59.8% improvement compared with conventional AlGaN/GaN HEMT. In addition, the average E–field in the region between gate and drain improves from 1.5 to 2.5 MV/cm. Based on the equivalent potential method (EPM), an analytical model of the E–field and potential distribution is presented. The veracity and effectiveness of the proposed methodology is verified by the good agreement between the simulated and modeled results.

The tradeoff between breakdown voltage (BV) and specific ON-resistance ( ${R}_{{\mathrm{\scriptscriptstyle ON},\text {sp}}}$ ) has always been the most crucial designing indicator to power devices. In this article, based on the effective concentration profile (ECP) concept, a novel field plate-adaptive doping (FAD) technique is proposed. Therefore, the benefits of the field plate (FP) structure and variation in lateral doping (VLD) technique are obtained simultaneously, while avoiding the inherent disadvantages of VLD and FP techniques. The new devices feature the gate FPs covering the part drift region with the FAD doping profile. The analytical model is therefore proposed to theoretically explore the breakdown mechanism. The good agreement between modeled and simulated results verifies the effectiveness of the FAD technique. The FAD achieves the advantages of low on-resistance and high BV by allowing a higher drift region doping concentration and suppressing the peak field at the FP side. Thus, compared with the FP and VLD, a better tradeoff between BV and ${R}_{{\mathrm{\scriptscriptstyle ON},\text {sp}}}$ can be obtained in the FAD lateral double-diffused metal–oxide–semiconductor (LDMOS).

Breakdown voltage (BV), on-state voltage (V_on), static latch-up voltage (V_lu), static latch-up current density (J_lu), and threshold voltage (V_th), etc., are critical static characteristic parameters of an IGBT for researchers. V_on and V_th can characterize the conduction capability of the device, while BV, V_lu, and J_lu can help designers analyze the safe operating area (SOA) of the device and its reliability. In this paper, we propose a multi-layer artificial neural network (ANN) framework to predict these characteristic parameters. The proposed scheme can accurately fit the relationship between structural parameters and static characteristic parameters. Given the structural parameters of the device, characteristic parameters can be generated accurately and efficiently. Compared with technology computer-aided design (TCAD) simulation, the average errors of our scheme for each characteristic parameter are within 8%, especially for BV and V_th, while the errors are controlled within 1%, and the evaluation speed is improved more than 10⁷ times. In addition, since the prediction process is mathematically a matrix operation process, there is no convergence problem, which there is in a TCAD simulation.